ONTAP Discussions

Moving from 7-Mode stretched MetroCluster to CDot MetroCluster

NICOLAS_MELAY
4,971 Views

Hi all.

 

We're about to replace our old FAS3140 MetroCluster (streched, single location) with a brand new FAS8040 CDot MetroCluster, half on premises and half hosted.

Moving from a dual controller 7-Mode MC to a 4 nodes CDot MC raises a number of questions.

 

Having a dual controller box on each location somewhat makes sense, and I guess NetApp had to go this way to fend off arguments about the alleged lack of local resiliency of 7-Mode fabric MetroClusters.

 

Howerver, what really annoys me is the need to split our storage capacity among the 4 nodes.

I understand that it's still better than a few months ago, when each node had to use a dedicated pair of shelves, and the minimum setup for a CDot MC was 4 FAS8040 nodes and 8 DS4246 shelves (plus 4 FC switches, 4 FireBridges, and so on).

Now we've been told that shelves can be "shared" among nodes, and the minmum setup has been lowered to just 4 nodes and 4 shelves.

Still, this setup implies the use of small aggregates (~10 drives) and many dedicated spare drives. (I've read that CDot wants 2 spare drives per node for garage something ?!?)

 

Really, is there no way to avoid splitting the storage pool in 4 quarters ?

Can't the secondary node on each site be put in a standby mode, without dedicated storage ?

Can't we use "root-data partitionning", like is being done on FAS2500s ?  (http://community.netapp.com/t5/Technology/Clustered-ONTAP-8-3-amp-Advanced-Drive-Partitioning/ba-p/98529)

Or is it coming anytime soon ?

And can't spare drives be shared among nodes ?

1 ACCEPTED SOLUTION

wenzel
4,928 Views

Disk Partitioning is not supported in MetroCluster for clustered Data ONTAP.

 

You do not need to assign the storage equally to the nodes. However, each node must have one root aggregate.

By default 2 non-root mirrored data aggregates are required per cluster, but since shortly this can be lowered to 1 non-root mirrored data aggregate per cluster. Please find more information on how to configure this here:

https://kb.netapp.com/support/index?page=content&id=1015185&locale=en_US

 

View solution in original post

3 REPLIES 3

DREW_RUSSELL
4,961 Views

I have extremly limited experience with Metrocluster but I can address some of your questions.

 

Can't we use "root-data partitionning", like is being done on FAS2500s

 

Advanced Drive Partiioning (ADP) is not availble on the FAS80XX line of controllers. Since the FAS80XX will typically be larger systems there is no need for ADP. If you have the bare minimum of 2 (fully populated) disk shelves on each side you're going have 48 disks total with 6 of those being dedicated the root parition on each node.

 

 

Still, this setup implies the use of small aggregates (~10 drives) and many dedicated spare drives. (I've read that CDot wants 2 spare drives per node for garage something ?!?)

 

Can you explain your reasoning behind the use of small aggregates here? What would be stopping you from increasing your aggregate size? As for your question in regards to clustered Data ONTAP wanting two spare drives per node, that is a new one for me. You may be refering to Flash storage and garbage collection, which is a method of extending the life of a Flash disk. I havent dealt much with the AFF either so I can be sure. 

wenzel
4,929 Views

Disk Partitioning is not supported in MetroCluster for clustered Data ONTAP.

 

You do not need to assign the storage equally to the nodes. However, each node must have one root aggregate.

By default 2 non-root mirrored data aggregates are required per cluster, but since shortly this can be lowered to 1 non-root mirrored data aggregate per cluster. Please find more information on how to configure this here:

https://kb.netapp.com/support/index?page=content&id=1015185&locale=en_US

 

NICOLAS_MELAY
4,500 Views

We ended up keeping dedicated aggregates on each node, since the unified cluster view conbined with datamotion for volumes makes it not so painful to manage.

Now I just wish our backup software had support for SVM-scoped NDMP...

Public